net/vmw_vsock/af_vsock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 /* Implementation notes:
   9  *
  10  * - There are two kinds of sockets: those created by user action (such as
  11  * calling socket(2)) and those created by incoming connection request packets.
  12  *
  13  * - There are two "global" tables, one for bound sockets (sockets that have
  14  * specified an address that they are responsible for) and one for connected
  15  * sockets (sockets that have established a connection with another socket).
  16  * These tables are "global" in that all sockets on the system are placed
  17  * within them. - Note, though, that the bound table contains an extra entry
  18  * for a list of unbound sockets and SOCK_DGRAM sockets will always remain in
  19  * that list. The bound table is used solely for lookup of sockets when packets
  20  * are received and that's not necessary for SOCK_DGRAM sockets since we create
  21  * a datagram handle for each and need not perform a lookup.  Keeping SOCK_DGRAM
  22  * sockets out of the bound hash buckets will reduce the chance of collisions
  23  * when looking for SOCK_STREAM sockets and prevents us from having to check the
  24  * socket type in the hash table lookups.
  25  *
  26  * - Sockets created by user action will either be "client" sockets that
  27  * initiate a connection or "server" sockets that listen for connections; we do
  28  * not support simultaneous connects (two "client" sockets connecting).
  29  *
  30  * - "Server" sockets are referred to as listener sockets throughout this
  31  * implementation because they are in the TCP_LISTEN state.  When a
  32  * connection request is received (the second kind of socket mentioned above),
  33  * we create a new socket and refer to it as a pending socket.  These pending
  34  * sockets are placed on the pending connection list of the listener socket.
  35  * When future packets are received for the address the listener socket is
  36  * bound to, we check if the source of the packet is from one that has an
  37  * existing pending connection.  If it does, we process the packet for the
  38  * pending socket.  When that socket reaches the connected state, it is removed
  39  * from the listener socket's pending list and enqueued in the listener
  40  * socket's accept queue.  Callers of accept(2) will accept connected sockets
  41  * from the listener socket's accept queue.  If the socket cannot be accepted
  42  * for some reason then it is marked rejected.  Once the connection is
  43  * accepted, it is owned by the user process and the responsibility for cleanup
  44  * falls with that user process.
  45  *
  46  * - It is possible that these pending sockets will never reach the connected
  47  * state; in fact, we may never receive another packet after the connection
  48  * request.  Because of this, we must schedule a cleanup function to run in the
  49  * future, after some amount of time passes where a connection should have been
  50  * established.  This function ensures that the socket is off all lists so it
  51  * cannot be retrieved, then drops all references to the socket so it is cleaned
  52  * up (sock_put() -> sk_free() -> our sk_destruct implementation).  Note this
  53  * function will also cleanup rejected sockets, those that reach the connected
  54  * state but leave it before they have been accepted.
  55  *
  56  * - Lock ordering for pending or accept queue sockets is:
  57  *
  58  *     lock_sock(listener);
  59  *     lock_sock_nested(pending, SINGLE_DEPTH_NESTING);
  60  *
  61  * Using explicit nested locking keeps lockdep happy since normally only one
  62  * lock of a given class may be taken at a time.
  63  *
  64  * - Sockets created by user action will be cleaned up when the user process
  65  * calls close(2), causing our release implementation to be called. Our release
  66  * implementation will perform some cleanup then drop the last reference so our
  67  * sk_destruct implementation is invoked.  Our sk_destruct implementation will
  68  * perform additional cleanup that's common for both types of sockets.
  69  *
  70  * - A socket's reference count is what ensures that the structure won't be
  71  * freed.  Each entry in a list (such as the "global" bound and connected tables
  72  * and the listener socket's pending list and connected queue) ensures a
  73  * reference.  When we defer work until process context and pass a socket as our
  74  * argument, we must ensure the reference count is increased to ensure the
  75  * socket isn't freed before the function is run; the deferred function will
  76  * then drop the reference.
  77  *
  78  * - sk->sk_state uses the TCP state constants because they are widely used by
  79  * other address families and exposed to userspace tools like ss(8):
  80  *
  81  *   TCP_CLOSE - unconnected
  82  *   TCP_SYN_SENT - connecting
  83  *   TCP_ESTABLISHED - connected
  84  *   TCP_CLOSING - disconnecting
  85  *   TCP_LISTEN - listening
  86  */
  87
  88 #include <linux/compat.h>
  89 #include <linux/types.h>
  90 #include <linux/bitops.h>
  91 #include <linux/cred.h>
  92 #include <linux/init.h>
  93 #include <linux/io.h>
  94 #include <linux/kernel.h>
  95 #include <linux/sched/signal.h>
  96 #include <linux/kmod.h>
  97 #include <linux/list.h>
  98 #include <linux/miscdevice.h>
  99 #include <linux/module.h>
 100 #include <linux/mutex.h>
 101 #include <linux/net.h>
 102 #include <linux/poll.h>
 103 #include <linux/random.h>
 104 #include <linux/skbuff.h>
 105 #include <linux/smp.h>
 106 #include <linux/socket.h>
 107 #include <linux/stddef.h>
 108 #include <linux/unistd.h>
 109 #include <linux/wait.h>
 110 #include <linux/workqueue.h>
 111 #include <net/sock.h>
 112 #include <net/af_vsock.h>
 113
 114 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr);
 115 static void vsock_sk_destruct(struct sock *sk);
 116 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 117
 118 /* Protocol family. */
 119 static struct proto vsock_proto = {
 120         .name = "AF_VSOCK",
 121         .owner = THIS_MODULE,
 122         .obj_size = sizeof(struct vsock_sock),
 123 };
 124
 125 /* The default peer timeout indicates how long we will wait for a peer response
 126  * to a control message.
 127  */
 128 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
 129
 130 #define VSOCK_DEFAULT_BUFFER_SIZE     (1024 * 256)
 131 #define VSOCK_DEFAULT_BUFFER_MAX_SIZE (1024 * 256)
 132 #define VSOCK_DEFAULT_BUFFER_MIN_SIZE 128
 133
 134 /* Transport used for host->guest communication */
 135 static const struct vsock_transport *transport_h2g;
 136 /* Transport used for guest->host communication */
 137 static const struct vsock_transport *transport_g2h;
 138 /* Transport used for DGRAM communication */
 139 static const struct vsock_transport *transport_dgram;
 140 /* Transport used for local communication */
 141 static const struct vsock_transport *transport_local;
 142 static DEFINE_MUTEX(vsock_register_mutex);
 143
 144 /**** UTILS ****/
 145
 146 /* Each bound VSocket is stored in the bind hash table and each connected
 147  * VSocket is stored in the connected hash table.
 148  *
 149  * Unbound sockets are all put on the same list attached to the end of the hash
 150  * table (vsock_unbound_sockets).  Bound sockets are added to the hash table in
 151  * the bucket that their local address hashes to (vsock_bound_sockets(addr)
 152  * represents the list that addr hashes to).
 153  *
 154  * Specifically, we initialize the vsock_bind_table array to a size of
 155  * VSOCK_HASH_SIZE + 1 so that vsock_bind_table[0] through
 156  * vsock_bind_table[VSOCK_HASH_SIZE - 1] are for bound sockets and
 157  * vsock_bind_table[VSOCK_HASH_SIZE] is for unbound sockets.  The hash function
 158  * mods with VSOCK_HASH_SIZE to ensure this.
 159  */
 160 #define MAX_PORT_RETRIES        24
 161
 162 #define VSOCK_HASH(addr)        ((addr)->svm_port % VSOCK_HASH_SIZE)
 163 #define vsock_bound_sockets(addr) (&vsock_bind_table[VSOCK_HASH(addr)])
 164 #define vsock_unbound_sockets     (&vsock_bind_table[VSOCK_HASH_SIZE])
 165
 166 /* XXX This can probably be implemented in a better way. */
 167 #define VSOCK_CONN_HASH(src, dst)                               \
 168         (((src)->svm_cid ^ (dst)->svm_port) % VSOCK_HASH_SIZE)
 169 #define vsock_connected_sockets(src, dst)               \
 170         (&vsock_connected_table[VSOCK_CONN_HASH(src, dst)])
 171 #define vsock_connected_sockets_vsk(vsk)                                \
 172         vsock_connected_sockets(&(vsk)->remote_addr, &(vsk)->local_addr)
 173
 174 struct list_head vsock_bind_table[VSOCK_HASH_SIZE + 1];
 175 EXPORT_SYMBOL_GPL(vsock_bind_table);
 176 struct list_head vsock_connected_table[VSOCK_HASH_SIZE];
 177 EXPORT_SYMBOL_GPL(vsock_connected_table);
 178 DEFINE_SPINLOCK(vsock_table_lock);
 179 EXPORT_SYMBOL_GPL(vsock_table_lock);
 180
 181 /* Autobind this socket to the local address if necessary. */
 182 static int vsock_auto_bind(struct vsock_sock *vsk)
 183 {
 184         struct sock *sk = sk_vsock(vsk);
 185         struct sockaddr_vm local_addr;
 186
 187         if (vsock_addr_bound(&vsk->local_addr))
 188                 return 0;
 189         vsock_addr_init(&local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 190         return __vsock_bind(sk, &local_addr);
 191 }
 192
 193 static void vsock_init_tables(void)
 194 {
 195         int i;
 196
 197         for (i = 0; i < ARRAY_SIZE(vsock_bind_table); i++)
 198                 INIT_LIST_HEAD(&vsock_bind_table[i]);
 199
 200         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++)
 201                 INIT_LIST_HEAD(&vsock_connected_table[i]);
 202 }
 203
 204 static void __vsock_insert_bound(struct list_head *list,
 205                                  struct vsock_sock *vsk)
 206 {
 207         sock_hold(&vsk->sk);
 208         list_add(&vsk->bound_table, list);
 209 }
 210
 211 static void __vsock_insert_connected(struct list_head *list,
 212                                      struct vsock_sock *vsk)
 213 {
 214         sock_hold(&vsk->sk);
 215         list_add(&vsk->connected_table, list);
 216 }
 217
 218 static void __vsock_remove_bound(struct vsock_sock *vsk)
 219 {
 220         list_del_init(&vsk->bound_table);
 221         sock_put(&vsk->sk);
 222 }
 223
 224 static void __vsock_remove_connected(struct vsock_sock *vsk)
 225 {
 226         list_del_init(&vsk->connected_table);
 227         sock_put(&vsk->sk);
 228 }
 229
 230 static struct sock *__vsock_find_bound_socket(struct sockaddr_vm *addr)
 231 {
 232         struct vsock_sock *vsk;
 233
 234         list_for_each_entry(vsk, vsock_bound_sockets(addr), bound_table) {
 235                 if (vsock_addr_equals_addr(addr, &vsk->local_addr))
 236                         return sk_vsock(vsk);
 237
 238                 if (addr->svm_port == vsk->local_addr.svm_port &&
 239                     (vsk->local_addr.svm_cid == VMADDR_CID_ANY ||
 240                      addr->svm_cid == VMADDR_CID_ANY))
 241                         return sk_vsock(vsk);
 242         }
 243
 244         return NULL;
 245 }
 246
 247 static struct sock *__vsock_find_connected_socket(struct sockaddr_vm *src,
 248                                                   struct sockaddr_vm *dst)
 249 {
 250         struct vsock_sock *vsk;
 251
 252         list_for_each_entry(vsk, vsock_connected_sockets(src, dst),
 253                             connected_table) {
 254                 if (vsock_addr_equals_addr(src, &vsk->remote_addr) &&
 255                     dst->svm_port == vsk->local_addr.svm_port) {
 256                         return sk_vsock(vsk);
 257                 }
 258         }
 259
 260         return NULL;
 261 }
 262
 263 static void vsock_insert_unbound(struct vsock_sock *vsk)
 264 {
 265         spin_lock_bh(&vsock_table_lock);
 266         __vsock_insert_bound(vsock_unbound_sockets, vsk);
 267         spin_unlock_bh(&vsock_table_lock);
 268 }
 269
 270 void vsock_insert_connected(struct vsock_sock *vsk)
 271 {
 272         struct list_head *list = vsock_connected_sockets(
 273                 &vsk->remote_addr, &vsk->local_addr);
 274
 275         spin_lock_bh(&vsock_table_lock);
 276         __vsock_insert_connected(list, vsk);
 277         spin_unlock_bh(&vsock_table_lock);
 278 }
 279 EXPORT_SYMBOL_GPL(vsock_insert_connected);
 280
 281 void vsock_remove_bound(struct vsock_sock *vsk)
 282 {
 283         spin_lock_bh(&vsock_table_lock);
 284         if (__vsock_in_bound_table(vsk))
 285                 __vsock_remove_bound(vsk);
 286         spin_unlock_bh(&vsock_table_lock);
 287 }
 288 EXPORT_SYMBOL_GPL(vsock_remove_bound);
 289
 290 void vsock_remove_connected(struct vsock_sock *vsk)
 291 {
 292         spin_lock_bh(&vsock_table_lock);
 293         if (__vsock_in_connected_table(vsk))
 294                 __vsock_remove_connected(vsk);
 295         spin_unlock_bh(&vsock_table_lock);
 296 }
 297 EXPORT_SYMBOL_GPL(vsock_remove_connected);
 298
 299 struct sock *vsock_find_bound_socket(struct sockaddr_vm *addr)
 300 {
 301         struct sock *sk;
 302
 303         spin_lock_bh(&vsock_table_lock);
 304         sk = __vsock_find_bound_socket(addr);
 305         if (sk)
 306                 sock_hold(sk);
 307
 308         spin_unlock_bh(&vsock_table_lock);
 309
 310         return sk;
 311 }
 312 EXPORT_SYMBOL_GPL(vsock_find_bound_socket);
 313
 314 struct sock *vsock_find_connected_socket(struct sockaddr_vm *src,
 315                                          struct sockaddr_vm *dst)
 316 {
 317         struct sock *sk;
 318
 319         spin_lock_bh(&vsock_table_lock);
 320         sk = __vsock_find_connected_socket(src, dst);
 321         if (sk)
 322                 sock_hold(sk);
 323
 324         spin_unlock_bh(&vsock_table_lock);
 325
 326         return sk;
 327 }
 328 EXPORT_SYMBOL_GPL(vsock_find_connected_socket);
 329
 330 void vsock_remove_sock(struct vsock_sock *vsk)
 331 {
 332         vsock_remove_bound(vsk);
 333         vsock_remove_connected(vsk);
 334 }
 335 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 336
 337 void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
 338 {
 339         int i;
 340
 341         spin_lock_bh(&vsock_table_lock);
 342
 343         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
 344                 struct vsock_sock *vsk;
 345                 list_for_each_entry(vsk, &vsock_connected_table[i],
 346                                     connected_table)
 347                         fn(sk_vsock(vsk));
 348         }
 349
 350         spin_unlock_bh(&vsock_table_lock);
 351 }
 352 EXPORT_SYMBOL_GPL(vsock_for_each_connected_socket);
 353
 354 void vsock_add_pending(struct sock *listener, struct sock *pending)
 355 {
 356         struct vsock_sock *vlistener;
 357         struct vsock_sock *vpending;
 358
 359         vlistener = vsock_sk(listener);
 360         vpending = vsock_sk(pending);
 361
 362         sock_hold(pending);
 363         sock_hold(listener);
 364         list_add_tail(&vpending->pending_links, &vlistener->pending_links);
 365 }
 366 EXPORT_SYMBOL_GPL(vsock_add_pending);
 367
 368 void vsock_remove_pending(struct sock *listener, struct sock *pending)
 369 {
 370         struct vsock_sock *vpending = vsock_sk(pending);
 371
 372         list_del_init(&vpending->pending_links);
 373         sock_put(listener);
 374         sock_put(pending);
 375 }
 376 EXPORT_SYMBOL_GPL(vsock_remove_pending);
 377
 378 void vsock_enqueue_accept(struct sock *listener, struct sock *connected)
 379 {
 380         struct vsock_sock *vlistener;
 381         struct vsock_sock *vconnected;
 382
 383         vlistener = vsock_sk(listener);
 384         vconnected = vsock_sk(connected);
 385
 386         sock_hold(connected);
 387         sock_hold(listener);
 388         list_add_tail(&vconnected->accept_queue, &vlistener->accept_queue);
 389 }
 390 EXPORT_SYMBOL_GPL(vsock_enqueue_accept);
 391
 392 static bool vsock_use_local_transport(unsigned int remote_cid)
 393 {
 394         if (!transport_local)
 395                 return false;
 396
 397         if (remote_cid == VMADDR_CID_LOCAL)
 398                 return true;
 399
 400         if (transport_g2h) {
 401                 return remote_cid == transport_g2h->get_local_cid();
 402         } else {
 403                 return remote_cid == VMADDR_CID_HOST;
 404         }
 405 }
 406
 407 static void vsock_deassign_transport(struct vsock_sock *vsk)
 408 {
 409         if (!vsk->transport)
 410                 return;
 411
 412         vsk->transport->destruct(vsk);
 413         module_put(vsk->transport->module);
 414         vsk->transport = NULL;
 415 }
 416
 417 /* Assign a transport to a socket and call the .init transport callback.
 418  *
 419  * Note: for connection oriented socket this must be called when vsk->remote_addr
 420  * is set (e.g. during the connect() or when a connection request on a listener
 421  * socket is received).
 422  * The vsk->remote_addr is used to decide which transport to use:
 423  *  - remote CID == VMADDR_CID_LOCAL or g2h->local_cid or VMADDR_CID_HOST if
 424  *    g2h is not loaded, will use local transport;
 425  *  - remote CID <= VMADDR_CID_HOST or h2g is not loaded or remote flags field
 426  *    includes VMADDR_FLAG_TO_HOST flag value, will use guest->host transport;
 427  *  - remote CID > VMADDR_CID_HOST will use host->guest transport;
 428  */
 429 int vsock_assign_transport(struct vsock_sock *vsk, struct vsock_sock *psk)
 430 {
 431         const struct vsock_transport *new_transport;
 432         struct sock *sk = sk_vsock(vsk);
 433         unsigned int remote_cid = vsk->remote_addr.svm_cid;
 434         __u8 remote_flags;
 435         int ret;
 436
 437         /* If the packet is coming with the source and destination CIDs higher
 438          * than VMADDR_CID_HOST, then a vsock channel where all the packets are
 439          * forwarded to the host should be established. Then the host will
 440          * need to forward the packets to the guest.
 441          *
 442          * The flag is set on the (listen) receive path (psk is not NULL). On
 443          * the connect path the flag can be set by the user space application.
 444          */
 445         if (psk && vsk->local_addr.svm_cid > VMADDR_CID_HOST &&
 446             vsk->remote_addr.svm_cid > VMADDR_CID_HOST)
 447                 vsk->remote_addr.svm_flags |= VMADDR_FLAG_TO_HOST;
 448
 449         remote_flags = vsk->remote_addr.svm_flags;
 450
 451         switch (sk->sk_type) {
 452         case SOCK_DGRAM:
 453                 new_transport = transport_dgram;
 454                 break;
 455         case SOCK_STREAM:
 456         case SOCK_SEQPACKET:
 457                 if (vsock_use_local_transport(remote_cid))
 458                         new_transport = transport_local;
 459                 else if (remote_cid <= VMADDR_CID_HOST || !transport_h2g ||
 460                          (remote_flags & VMADDR_FLAG_TO_HOST))
 461                         new_transport = transport_g2h;
 462                 else
 463                         new_transport = transport_h2g;
 464                 break;
 465         default:
 466                 return -ESOCKTNOSUPPORT;
 467         }
 468
 469         if (vsk->transport) {
 470                 if (vsk->transport == new_transport)
 471                         return 0;
 472
 473                 /* transport->release() must be called with sock lock acquired.
 474                  * This path can only be taken during vsock_connect(), where we
 475                  * have already held the sock lock. In the other cases, this
 476                  * function is called on a new socket which is not assigned to
 477                  * any transport.
 478                  */
 479                 vsk->transport->release(vsk);
 480                 vsock_deassign_transport(vsk);
 481         }
 482
 483         /* We increase the module refcnt to prevent the transport unloading
 484          * while there are open sockets assigned to it.
 485          */
 486         if (!new_transport || !try_module_get(new_transport->module))
 487                 return -ENODEV;
 488
 489         if (sk->sk_type == SOCK_SEQPACKET) {
 490                 if (!new_transport->seqpacket_allow ||
 491                     !new_transport->seqpacket_allow(remote_cid)) {
 492                         module_put(new_transport->module);
 493                         return -ESOCKTNOSUPPORT;
 494                 }
 495         }
 496
 497         ret = new_transport->init(vsk, psk);
 498         if (ret) {
 499                 module_put(new_transport->module);
 500                 return ret;
 501         }
 502
 503         vsk->transport = new_transport;
 504
 505         return 0;
 506 }
 507 EXPORT_SYMBOL_GPL(vsock_assign_transport);
 508
 509 bool vsock_find_cid(unsigned int cid)
 510 {
 511         if (transport_g2h && cid == transport_g2h->get_local_cid())
 512                 return true;
 513
 514         if (transport_h2g && cid == VMADDR_CID_HOST)
 515                 return true;
 516
 517         if (transport_local && cid == VMADDR_CID_LOCAL)
 518                 return true;
 519
 520         return false;
 521 }
 522 EXPORT_SYMBOL_GPL(vsock_find_cid);
 523
 524 static struct sock *vsock_dequeue_accept(struct sock *listener)
 525 {
 526         struct vsock_sock *vlistener;
 527         struct vsock_sock *vconnected;
 528
 529         vlistener = vsock_sk(listener);
 530
 531         if (list_empty(&vlistener->accept_queue))
 532                 return NULL;
 533
 534         vconnected = list_entry(vlistener->accept_queue.next,
 535                                 struct vsock_sock, accept_queue);
 536
 537         list_del_init(&vconnected->accept_queue);
 538         sock_put(listener);
 539         /* The caller will need a reference on the connected socket so we let
 540          * it call sock_put().
 541          */
 542
 543         return sk_vsock(vconnected);
 544 }
 545
 546 static bool vsock_is_accept_queue_empty(struct sock *sk)
 547 {
 548         struct vsock_sock *vsk = vsock_sk(sk);
 549         return list_empty(&vsk->accept_queue);
 550 }
 551
 552 static bool vsock_is_pending(struct sock *sk)
 553 {
 554         struct vsock_sock *vsk = vsock_sk(sk);
 555         return !list_empty(&vsk->pending_links);
 556 }
 557
 558 static int vsock_send_shutdown(struct sock *sk, int mode)
 559 {
 560         struct vsock_sock *vsk = vsock_sk(sk);
 561
 562         if (!vsk->transport)
 563                 return -ENODEV;
 564
 565         return vsk->transport->shutdown(vsk, mode);
 566 }
 567
 568 static void vsock_pending_work(struct work_struct *work)
 569 {
 570         struct sock *sk;
 571         struct sock *listener;
 572         struct vsock_sock *vsk;
 573         bool cleanup;
 574
 575         vsk = container_of(work, struct vsock_sock, pending_work.work);
 576         sk = sk_vsock(vsk);
 577         listener = vsk->listener;
 578         cleanup = true;
 579
 580         lock_sock(listener);
 581         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 582
 583         if (vsock_is_pending(sk)) {
 584                 vsock_remove_pending(listener, sk);
 585
 586                 sk_acceptq_removed(listener);
 587         } else if (!vsk->rejected) {
 588                 /* We are not on the pending list and accept() did not reject
 589                  * us, so we must have been accepted by our user process.  We
 590                  * just need to drop our references to the sockets and be on
 591                  * our way.
 592                  */
 593                 cleanup = false;
 594                 goto out;
 595         }
 596
 597         /* We need to remove ourself from the global connected sockets list so
 598          * incoming packets can't find this socket, and to reduce the reference
 599          * count.
 600          */
 601         vsock_remove_connected(vsk);
 602
 603         sk->sk_state = TCP_CLOSE;
 604
 605 out:
 606         release_sock(sk);
 607         release_sock(listener);
 608         if (cleanup)
 609                 sock_put(sk);
 610
 611         sock_put(sk);
 612         sock_put(listener);
 613 }
 614
 615 /**** SOCKET OPERATIONS ****/
 616
 617 static int __vsock_bind_connectible(struct vsock_sock *vsk,
 618                                     struct sockaddr_vm *addr)
 619 {
 620         static u32 port;
 621         struct sockaddr_vm new_addr;
 622
 623         if (!port)
 624                 port = LAST_RESERVED_PORT + 1 +
 625                         prandom_u32_max(U32_MAX - LAST_RESERVED_PORT);
 626
 627         vsock_addr_init(&new_addr, addr->svm_cid, addr->svm_port);
 628
 629         if (addr->svm_port == VMADDR_PORT_ANY) {
 630                 bool found = false;
 631                 unsigned int i;
 632
 633                 for (i = 0; i < MAX_PORT_RETRIES; i++) {
 634                         if (port <= LAST_RESERVED_PORT)
 635                                 port = LAST_RESERVED_PORT + 1;
 636
 637                         new_addr.svm_port = port++;
 638
 639                         if (!__vsock_find_bound_socket(&new_addr)) {
 640                                 found = true;
 641                                 break;
 642                         }
 643                 }
 644
 645                 if (!found)
 646                         return -EADDRNOTAVAIL;
 647         } else {
 648                 /* If port is in reserved range, ensure caller
 649                  * has necessary privileges.
 650                  */
 651                 if (addr->svm_port <= LAST_RESERVED_PORT &&
 652                     !capable(CAP_NET_BIND_SERVICE)) {
 653                         return -EACCES;
 654                 }
 655
 656                 if (__vsock_find_bound_socket(&new_addr))
 657                         return -EADDRINUSE;
 658         }
 659
 660         vsock_addr_init(&vsk->local_addr, new_addr.svm_cid, new_addr.svm_port);
 661
 662         /* Remove connection oriented sockets from the unbound list and add them
 663          * to the hash table for easy lookup by its address.  The unbound list
 664          * is simply an extra entry at the end of the hash table, a trick used
 665          * by AF_UNIX.
 666          */
 667         __vsock_remove_bound(vsk);
 668         __vsock_insert_bound(vsock_bound_sockets(&vsk->local_addr), vsk);
 669
 670         return 0;
 671 }
 672
 673 static int __vsock_bind_dgram(struct vsock_sock *vsk,
 674                               struct sockaddr_vm *addr)
 675 {
 676         return vsk->transport->dgram_bind(vsk, addr);
 677 }
 678
 679 static int __vsock_bind(struct sock *sk, struct sockaddr_vm *addr)
 680 {
 681         struct vsock_sock *vsk = vsock_sk(sk);
 682         int retval;
 683
 684         /* First ensure this socket isn't already bound. */
 685         if (vsock_addr_bound(&vsk->local_addr))
 686                 return -EINVAL;
 687
 688         /* Now bind to the provided address or select appropriate values if
 689          * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
 690          * like AF_INET prevents binding to a non-local IP address (in most
 691          * cases), we only allow binding to a local CID.
 692          */
 693         if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
 694                 return -EADDRNOTAVAIL;
 695
 696         switch (sk->sk_socket->type) {
 697         case SOCK_STREAM:
 698         case SOCK_SEQPACKET:
 699                 spin_lock_bh(&vsock_table_lock);
 700                 retval = __vsock_bind_connectible(vsk, addr);
 701                 spin_unlock_bh(&vsock_table_lock);
 702                 break;
 703
 704         case SOCK_DGRAM:
 705                 retval = __vsock_bind_dgram(vsk, addr);
 706                 break;
 707
 708         default:
 709                 retval = -EINVAL;
 710                 break;
 711         }
 712
 713         return retval;
 714 }
 715
 716 static void vsock_connect_timeout(struct work_struct *work);
 717
 718 static struct sock *__vsock_create(struct net *net,
 719                                    struct socket *sock,
 720                                    struct sock *parent,
 721                                    gfp_t priority,
 722                                    unsigned short type,
 723                                    int kern)
 724 {
 725         struct sock *sk;
 726         struct vsock_sock *psk;
 727         struct vsock_sock *vsk;
 728
 729         sk = sk_alloc(net, AF_VSOCK, priority, &vsock_proto, kern);
 730         if (!sk)
 731                 return NULL;
 732
 733         sock_init_data(sock, sk);
 734
 735         /* sk->sk_type is normally set in sock_init_data, but only if sock is
 736          * non-NULL. We make sure that our sockets always have a type by
 737          * setting it here if needed.
 738          */
 739         if (!sock)
 740                 sk->sk_type = type;
 741
 742         vsk = vsock_sk(sk);
 743         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 744         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 745
 746         sk->sk_destruct = vsock_sk_destruct;
 747         sk->sk_backlog_rcv = vsock_queue_rcv_skb;
 748         sock_reset_flag(sk, SOCK_DONE);
 749
 750         INIT_LIST_HEAD(&vsk->bound_table);
 751         INIT_LIST_HEAD(&vsk->connected_table);
 752         vsk->listener = NULL;
 753         INIT_LIST_HEAD(&vsk->pending_links);
 754         INIT_LIST_HEAD(&vsk->accept_queue);
 755         vsk->rejected = false;
 756         vsk->sent_request = false;
 757         vsk->ignore_connecting_rst = false;
 758         vsk->peer_shutdown = 0;
 759         INIT_DELAYED_WORK(&vsk->connect_work, vsock_connect_timeout);
 760         INIT_DELAYED_WORK(&vsk->pending_work, vsock_pending_work);
 761
 762         psk = parent ? vsock_sk(parent) : NULL;
 763         if (parent) {
 764                 vsk->trusted = psk->trusted;
 765                 vsk->owner = get_cred(psk->owner);
 766                 vsk->connect_timeout = psk->connect_timeout;
 767                 vsk->buffer_size = psk->buffer_size;
 768                 vsk->buffer_min_size = psk->buffer_min_size;
 769                 vsk->buffer_max_size = psk->buffer_max_size;
 770                 security_sk_clone(parent, sk);
 771         } else {
 772                 vsk->trusted = ns_capable_noaudit(&init_user_ns, CAP_NET_ADMIN);
 773                 vsk->owner = get_current_cred();
 774                 vsk->connect_timeout = VSOCK_DEFAULT_CONNECT_TIMEOUT;
 775                 vsk->buffer_size = VSOCK_DEFAULT_BUFFER_SIZE;
 776                 vsk->buffer_min_size = VSOCK_DEFAULT_BUFFER_MIN_SIZE;
 777                 vsk->buffer_max_size = VSOCK_DEFAULT_BUFFER_MAX_SIZE;
 778         }
 779
 780         return sk;
 781 }
 782
 783 static bool sock_type_connectible(u16 type)
 784 {
 785         return (type == SOCK_STREAM) || (type == SOCK_SEQPACKET);
 786 }
 787
 788 static void __vsock_release(struct sock *sk, int level)
 789 {
 790         if (sk) {
 791                 struct sock *pending;
 792                 struct vsock_sock *vsk;
 793
 794                 vsk = vsock_sk(sk);
 795                 pending = NULL; /* Compiler warning. */
 796
 797                 /* When "level" is SINGLE_DEPTH_NESTING, use the nested
 798                  * version to avoid the warning "possible recursive locking
 799                  * detected". When "level" is 0, lock_sock_nested(sk, level)
 800                  * is the same as lock_sock(sk).
 801                  */
 802                 lock_sock_nested(sk, level);
 803
 804                 if (vsk->transport)
 805                         vsk->transport->release(vsk);
 806                 else if (sock_type_connectible(sk->sk_type))
 807                         vsock_remove_sock(vsk);
 808
 809                 sock_orphan(sk);
 810                 sk->sk_shutdown = SHUTDOWN_MASK;
 811
 812                 skb_queue_purge(&sk->sk_receive_queue);
 813
 814                 /* Clean up any sockets that never were accepted. */
 815                 while ((pending = vsock_dequeue_accept(sk)) != NULL) {
 816                         __vsock_release(pending, SINGLE_DEPTH_NESTING);
 817                         sock_put(pending);
 818                 }
 819
 820                 release_sock(sk);
 821                 sock_put(sk);
 822         }
 823 }
 824
 825 static void vsock_sk_destruct(struct sock *sk)
 826 {
 827         struct vsock_sock *vsk = vsock_sk(sk);
 828
 829         vsock_deassign_transport(vsk);
 830
 831         /* When clearing these addresses, there's no need to set the family and
 832          * possibly register the address family with the kernel.
 833          */
 834         vsock_addr_init(&vsk->local_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 835         vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY, VMADDR_PORT_ANY);
 836
 837         put_cred(vsk->owner);
 838 }
 839
 840 static int vsock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 841 {
 842         int err;
 843
 844         err = sock_queue_rcv_skb(sk, skb);
 845         if (err)
 846                 kfree_skb(skb);
 847
 848         return err;
 849 }
 850
 851 struct sock *vsock_create_connected(struct sock *parent)
 852 {
 853         return __vsock_create(sock_net(parent), NULL, parent, GFP_KERNEL,
 854                               parent->sk_type, 0);
 855 }
 856 EXPORT_SYMBOL_GPL(vsock_create_connected);
 857
 858 s64 vsock_stream_has_data(struct vsock_sock *vsk)
 859 {
 860         return vsk->transport->stream_has_data(vsk);
 861 }
 862 EXPORT_SYMBOL_GPL(vsock_stream_has_data);
 863
 864 static s64 vsock_connectible_has_data(struct vsock_sock *vsk)
 865 {
 866         struct sock *sk = sk_vsock(vsk);
 867
 868         if (sk->sk_type == SOCK_SEQPACKET)
 869                 return vsk->transport->seqpacket_has_data(vsk);
 870         else
 871                 return vsock_stream_has_data(vsk);
 872 }
 873
 874 s64 vsock_stream_has_space(struct vsock_sock *vsk)
 875 {
 876         return vsk->transport->stream_has_space(vsk);
 877 }
 878 EXPORT_SYMBOL_GPL(vsock_stream_has_space);
 879
 880 static int vsock_release(struct socket *sock)
 881 {
 882         __vsock_release(sock->sk, 0);
 883         sock->sk = NULL;
 884         sock->state = SS_FREE;
 885
 886         return 0;
 887 }
 888
 889 static int
 890 vsock_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 891 {
 892         int err;
 893         struct sock *sk;
 894         struct sockaddr_vm *vm_addr;
 895
 896         sk = sock->sk;
 897
 898         if (vsock_addr_cast(addr, addr_len, &vm_addr) != 0)
 899                 return -EINVAL;
 900
 901         lock_sock(sk);
 902         err = __vsock_bind(sk, vm_addr);
 903         release_sock(sk);
 904
 905         return err;
 906 }
 907
 908 static int vsock_getname(struct socket *sock,
 909                          struct sockaddr *addr, int peer)
 910 {
 911         int err;
 912         struct sock *sk;
 913         struct vsock_sock *vsk;
 914         struct sockaddr_vm *vm_addr;
 915
 916         sk = sock->sk;
 917         vsk = vsock_sk(sk);
 918         err = 0;
 919
 920         lock_sock(sk);
 921
 922         if (peer) {
 923                 if (sock->state != SS_CONNECTED) {
 924                         err = -ENOTCONN;
 925                         goto out;
 926                 }
 927                 vm_addr = &vsk->remote_addr;
 928         } else {
 929                 vm_addr = &vsk->local_addr;
 930         }
 931
 932         if (!vm_addr) {
 933                 err = -EINVAL;
 934                 goto out;
 935         }
 936
 937         /* sys_getsockname() and sys_getpeername() pass us a
 938          * MAX_SOCK_ADDR-sized buffer and don't set addr_len.  Unfortunately
 939          * that macro is defined in socket.c instead of .h, so we hardcode its
 940          * value here.
 941          */
 942         BUILD_BUG_ON(sizeof(*vm_addr) > 128);
 943         memcpy(addr, vm_addr, sizeof(*vm_addr));
 944         err = sizeof(*vm_addr);
 945
 946 out:
 947         release_sock(sk);
 948         return err;
 949 }
 950
 951 static int vsock_shutdown(struct socket *sock, int mode)
 952 {
 953         int err;
 954         struct sock *sk;
 955
 956         /* User level uses SHUT_RD (0) and SHUT_WR (1), but the kernel uses
 957          * RCV_SHUTDOWN (1) and SEND_SHUTDOWN (2), so we must increment mode
 958          * here like the other address families do.  Note also that the
 959          * increment makes SHUT_RDWR (2) into RCV_SHUTDOWN | SEND_SHUTDOWN (3),
 960          * which is what we want.
 961          */
 962         mode++;
 963
 964         if ((mode & ~SHUTDOWN_MASK) || !mode)
 965                 return -EINVAL;
 966
 967         /* If this is a connection oriented socket and it is not connected then
 968          * bail out immediately.  If it is a DGRAM socket then we must first
 969          * kick the socket so that it wakes up from any sleeping calls, for
 970          * example recv(), and then afterwards return the error.
 971          */
 972
 973         sk = sock->sk;
 974
 975         lock_sock(sk);
 976         if (sock->state == SS_UNCONNECTED) {
 977                 err = -ENOTCONN;
 978                 if (sock_type_connectible(sk->sk_type))
 979                         goto out;
 980         } else {
 981                 sock->state = SS_DISCONNECTING;
 982                 err = 0;
 983         }
 984
 985         /* Receive and send shutdowns are treated alike. */
 986         mode = mode & (RCV_SHUTDOWN | SEND_SHUTDOWN);
 987         if (mode) {
 988                 sk->sk_shutdown |= mode;
 989                 sk->sk_state_change(sk);
 990
 991                 if (sock_type_connectible(sk->sk_type)) {
 992                         sock_reset_flag(sk, SOCK_DONE);
 993                         vsock_send_shutdown(sk, mode);
 994                 }
 995         }
 996
 997 out:
 998         release_sock(sk);
 999         return err;
1000 }
1001
1002 static __poll_t vsock_poll(struct file *file, struct socket *sock,
1003                                poll_table *wait)
1004 {
1005         struct sock *sk;
1006         __poll_t mask;
1007         struct vsock_sock *vsk;
1008
1009         sk = sock->sk;
1010         vsk = vsock_sk(sk);
1011
1012         poll_wait(file, sk_sleep(sk), wait);
1013         mask = 0;
1014
1015         if (sk->sk_err)
1016                 /* Signify that there has been an error on this socket. */
1017                 mask |= EPOLLERR;
1018
1019         /* INET sockets treat local write shutdown and peer write shutdown as a
1020          * case of EPOLLHUP set.
1021          */
1022         if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1023             ((sk->sk_shutdown & SEND_SHUTDOWN) &&
1024              (vsk->peer_shutdown & SEND_SHUTDOWN))) {
1025                 mask |= EPOLLHUP;
1026         }
1027
1028         if (sk->sk_shutdown & RCV_SHUTDOWN ||
1029             vsk->peer_shutdown & SEND_SHUTDOWN) {
1030                 mask |= EPOLLRDHUP;
1031         }
1032
1033         if (sock->type == SOCK_DGRAM) {
1034                 /* For datagram sockets we can read if there is something in
1035                  * the queue and write as long as the socket isn't shutdown for
1036                  * sending.
1037                  */
1038                 if (!skb_queue_empty_lockless(&sk->sk_receive_queue) ||
1039                     (sk->sk_shutdown & RCV_SHUTDOWN)) {
1040                         mask |= EPOLLIN | EPOLLRDNORM;
1041                 }
1042
1043                 if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1044                         mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
1045
1046         } else if (sock_type_connectible(sk->sk_type)) {
1047                 const struct vsock_transport *transport;
1048
1049                 lock_sock(sk);
1050
1051                 transport = vsk->transport;
1052
1053                 /* Listening sockets that have connections in their accept
1054                  * queue can be read.
1055                  */
1056                 if (sk->sk_state == TCP_LISTEN
1057                     && !vsock_is_accept_queue_empty(sk))
1058                         mask |= EPOLLIN | EPOLLRDNORM;
1059
1060                 /* If there is something in the queue then we can read. */
1061                 if (transport && transport->stream_is_active(vsk) &&
1062                     !(sk->sk_shutdown & RCV_SHUTDOWN)) {
1063                         bool data_ready_now = false;
1064                         int ret = transport->notify_poll_in(
1065                                         vsk, 1, &data_ready_now);
1066                         if (ret < 0) {
1067                                 mask |= EPOLLERR;
1068                         } else {
1069                                 if (data_ready_now)
1070                                         mask |= EPOLLIN | EPOLLRDNORM;
1071
1072                         }
1073                 }
1074
1075                 /* Sockets whose connections have been closed, reset, or
1076                  * terminated should also be considered read, and we check the
1077                  * shutdown flag for that.
1078                  */
1079                 if (sk->sk_shutdown & RCV_SHUTDOWN ||
1080                     vsk->peer_shutdown & SEND_SHUTDOWN) {
1081                         mask |= EPOLLIN | EPOLLRDNORM;
1082                 }
1083
1084                 /* Connected sockets that can produce data can be written. */
1085                 if (transport && sk->sk_state == TCP_ESTABLISHED) {
1086                         if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1087                                 bool space_avail_now = false;
1088                                 int ret = transport->notify_poll_out(
1089                                                 vsk, 1, &space_avail_now);
1090                                 if (ret < 0) {
1091                                         mask |= EPOLLERR;
1092                                 } else {
1093                                         if (space_avail_now)
1094                                                 /* Remove EPOLLWRBAND since INET
1095                                                  * sockets are not setting it.
1096                                                  */
1097                                                 mask |= EPOLLOUT | EPOLLWRNORM;
1098
1099                                 }
1100                         }
1101                 }
1102
1103                 /* Simulate INET socket poll behaviors, which sets
1104                  * EPOLLOUT|EPOLLWRNORM when peer is closed and nothing to read,
1105                  * but local send is not shutdown.
1106                  */
1107                 if (sk->sk_state == TCP_CLOSE || sk->sk_state == TCP_CLOSING) {
1108                         if (!(sk->sk_shutdown & SEND_SHUTDOWN))
1109                                 mask |= EPOLLOUT | EPOLLWRNORM;
1110
1111                 }
1112
1113                 release_sock(sk);
1114         }
1115
1116         return mask;
1117 }
1118
1119 static int vsock_dgram_sendmsg(struct socket *sock, struct msghdr *msg,
1120                                size_t len)
1121 {
1122         int err;
1123         struct sock *sk;
1124         struct vsock_sock *vsk;
1125         struct sockaddr_vm *remote_addr;
1126         const struct vsock_transport *transport;
1127
1128         if (msg->msg_flags & MSG_OOB)
1129                 return -EOPNOTSUPP;
1130
1131         /* For now, MSG_DONTWAIT is always assumed... */
1132         err = 0;
1133         sk = sock->sk;
1134         vsk = vsock_sk(sk);
1135
1136         lock_sock(sk);
1137
1138         transport = vsk->transport;
1139
1140         err = vsock_auto_bind(vsk);
1141         if (err)
1142                 goto out;
1143
1144
1145         /* If the provided message contains an address, use that.  Otherwise
1146          * fall back on the socket's remote handle (if it has been connected).
1147          */
1148         if (msg->msg_name &&
1149             vsock_addr_cast(msg->msg_name, msg->msg_namelen,
1150                             &remote_addr) == 0) {
1151                 /* Ensure this address is of the right type and is a valid
1152                  * destination.
1153                  */
1154
1155                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1156                         remote_addr->svm_cid = transport->get_local_cid();
1157
1158                 if (!vsock_addr_bound(remote_addr)) {
1159                         err = -EINVAL;
1160                         goto out;
1161                 }
1162         } else if (sock->state == SS_CONNECTED) {
1163                 remote_addr = &vsk->remote_addr;
1164
1165                 if (remote_addr->svm_cid == VMADDR_CID_ANY)
1166                         remote_addr->svm_cid = transport->get_local_cid();
1167
1168                 /* XXX Should connect() or this function ensure remote_addr is
1169                  * bound?
1170                  */
1171                 if (!vsock_addr_bound(&vsk->remote_addr)) {
1172                         err = -EINVAL;
1173                         goto out;
1174                 }
1175         } else {
1176                 err = -EINVAL;
1177                 goto out;
1178         }
1179
1180         if (!transport->dgram_allow(remote_addr->svm_cid,
1181                                     remote_addr->svm_port)) {
1182                 err = -EINVAL;
1183                 goto out;
1184         }
1185
1186         err = transport->dgram_enqueue(vsk, remote_addr, msg, len);
1187
1188 out:
1189         release_sock(sk);
1190         return err;
1191 }
1192
1193 static int vsock_dgram_connect(struct socket *sock,
1194                                struct sockaddr *addr, int addr_len, int flags)
1195 {
1196         int err;
1197         struct sock *sk;
1198         struct vsock_sock *vsk;
1199         struct sockaddr_vm *remote_addr;
1200
1201         sk = sock->sk;
1202         vsk = vsock_sk(sk);
1203
1204         err = vsock_addr_cast(addr, addr_len, &remote_addr);
1205         if (err == -EAFNOSUPPORT && remote_addr->svm_family == AF_UNSPEC) {
1206                 lock_sock(sk);
1207                 vsock_addr_init(&vsk->remote_addr, VMADDR_CID_ANY,
1208                                 VMADDR_PORT_ANY);
1209                 sock->state = SS_UNCONNECTED;
1210                 release_sock(sk);
1211                 return 0;
1212         } else if (err != 0)
1213                 return -EINVAL;
1214
1215         lock_sock(sk);
1216
1217         err = vsock_auto_bind(vsk);
1218         if (err)
1219                 goto out;
1220
1221         if (!vsk->transport->dgram_allow(remote_addr->svm_cid,
1222                                          remote_addr->svm_port)) {
1223                 err = -EINVAL;
1224                 goto out;
1225         }
1226
1227         memcpy(&vsk->remote_addr, remote_addr, sizeof(vsk->remote_addr));
1228         sock->state = SS_CONNECTED;
1229
1230 out:
1231         release_sock(sk);
1232         return err;
1233 }
1234
1235 static int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
1236                                size_t len, int flags)
1237 {
1238         struct vsock_sock *vsk = vsock_sk(sock->sk);
1239
1240         return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
1241 }
1242
1243 static const struct proto_ops vsock_dgram_ops = {
1244         .family = PF_VSOCK,
1245         .owner = THIS_MODULE,
1246         .release = vsock_release,
1247         .bind = vsock_bind,
1248         .connect = vsock_dgram_connect,
1249         .socketpair = sock_no_socketpair,
1250         .accept = sock_no_accept,
1251         .getname = vsock_getname,
1252         .poll = vsock_poll,
1253         .ioctl = sock_no_ioctl,
1254         .listen = sock_no_listen,
1255         .shutdown = vsock_shutdown,
1256         .sendmsg = vsock_dgram_sendmsg,
1257         .recvmsg = vsock_dgram_recvmsg,
1258         .mmap = sock_no_mmap,
1259         .sendpage = sock_no_sendpage,
1260 };
1261
1262 static int vsock_transport_cancel_pkt(struct vsock_sock *vsk)
1263 {
1264         const struct vsock_transport *transport = vsk->transport;
1265
1266         if (!transport || !transport->cancel_pkt)
1267                 return -EOPNOTSUPP;
1268
1269         return transport->cancel_pkt(vsk);
1270 }
1271
1272 static void vsock_connect_timeout(struct work_struct *work)
1273 {
1274         struct sock *sk;
1275         struct vsock_sock *vsk;
1276
1277         vsk = container_of(work, struct vsock_sock, connect_work.work);
1278         sk = sk_vsock(vsk);
1279
1280         lock_sock(sk);
1281         if (sk->sk_state == TCP_SYN_SENT &&
1282             (sk->sk_shutdown != SHUTDOWN_MASK)) {
1283                 sk->sk_state = TCP_CLOSE;
1284                 sk->sk_err = ETIMEDOUT;
1285                 sk_error_report(sk);
1286                 vsock_transport_cancel_pkt(vsk);
1287         }
1288         release_sock(sk);
1289
1290         sock_put(sk);
1291 }
1292
1293 static int vsock_connect(struct socket *sock, struct sockaddr *addr,
1294                          int addr_len, int flags)
1295 {
1296         int err;
1297         struct sock *sk;
1298         struct vsock_sock *vsk;
1299         const struct vsock_transport *transport;
1300         struct sockaddr_vm *remote_addr;
1301         long timeout;
1302         DEFINE_WAIT(wait);
1303
1304         err = 0;
1305         sk = sock->sk;
1306         vsk = vsock_sk(sk);
1307
1308         lock_sock(sk);
1309
1310         /* XXX AF_UNSPEC should make us disconnect like AF_INET. */
1311         switch (sock->state) {
1312         case SS_CONNECTED:
1313                 err = -EISCONN;
1314                 goto out;
1315         case SS_DISCONNECTING:
1316                 err = -EINVAL;
1317                 goto out;
1318         case SS_CONNECTING:
1319                 /* This continues on so we can move sock into the SS_CONNECTED
1320                  * state once the connection has completed (at which point err
1321                  * will be set to zero also).  Otherwise, we will either wait
1322                  * for the connection or return -EALREADY should this be a
1323                  * non-blocking call.
1324                  */
1325                 err = -EALREADY;
1326                 if (flags & O_NONBLOCK)
1327                         goto out;
1328                 break;
1329         default:
1330                 if ((sk->sk_state == TCP_LISTEN) ||
1331                     vsock_addr_cast(addr, addr_len, &remote_addr) != 0) {
1332                         err = -EINVAL;
1333                         goto out;
1334                 }
1335
1336                 /* Set the remote address that we are connecting to. */
1337                 memcpy(&vsk->remote_addr, remote_addr,
1338                        sizeof(vsk->remote_addr));
1339
1340                 err = vsock_assign_transport(vsk, NULL);
1341                 if (err)
1342                         goto out;
1343
1344                 transport = vsk->transport;
1345
1346                 /* The hypervisor and well-known contexts do not have socket
1347                  * endpoints.
1348                  */
1349                 if (!transport ||
1350                     !transport->stream_allow(remote_addr->svm_cid,
1351                                              remote_addr->svm_port)) {
1352                         err = -ENETUNREACH;
1353                         goto out;
1354                 }
1355
1356                 err = vsock_auto_bind(vsk);
1357                 if (err)
1358                         goto out;
1359
1360                 sk->sk_state = TCP_SYN_SENT;
1361
1362                 err = transport->connect(vsk);
1363                 if (err < 0)
1364                         goto out;
1365
1366                 /* Mark sock as connecting and set the error code to in
1367                  * progress in case this is a non-blocking connect.
1368                  */
1369                 sock->state = SS_CONNECTING;
1370                 err = -EINPROGRESS;
1371         }
1372
1373         /* The receive path will handle all communication until we are able to
1374          * enter the connected state.  Here we wait for the connection to be
1375          * completed or a notification of an error.
1376          */
1377         timeout = vsk->connect_timeout;
1378         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1379
1380         while (sk->sk_state != TCP_ESTABLISHED && sk->sk_err == 0) {
1381                 if (flags & O_NONBLOCK) {
1382                         /* If we're not going to block, we schedule a timeout
1383                          * function to generate a timeout on the connection
1384                          * attempt, in case the peer doesn't respond in a
1385                          * timely manner. We hold on to the socket until the
1386                          * timeout fires.
1387                          */
1388                         sock_hold(sk);
1389                         schedule_delayed_work(&vsk->connect_work, timeout);
1390
1391                         /* Skip ahead to preserve error code set above. */
1392                         goto out_wait;
1393                 }
1394
1395                 release_sock(sk);
1396                 timeout = schedule_timeout(timeout);
1397                 lock_sock(sk);
1398
1399                 if (signal_pending(current)) {
1400                         err = sock_intr_errno(timeout);
1401                         sk->sk_state = sk->sk_state == TCP_ESTABLISHED ? TCP_CLOSING : TCP_CLOSE;
1402                         sock->state = SS_UNCONNECTED;
1403                         vsock_transport_cancel_pkt(vsk);
1404                         vsock_remove_connected(vsk);
1405                         goto out_wait;
1406                 } else if (timeout == 0) {
1407                         err = -ETIMEDOUT;
1408                         sk->sk_state = TCP_CLOSE;
1409                         sock->state = SS_UNCONNECTED;
1410                         vsock_transport_cancel_pkt(vsk);
1411                         goto out_wait;
1412                 }
1413
1414                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1415         }
1416
1417         if (sk->sk_err) {
1418                 err = -sk->sk_err;
1419                 sk->sk_state = TCP_CLOSE;
1420                 sock->state = SS_UNCONNECTED;
1421         } else {
1422                 err = 0;
1423         }
1424
1425 out_wait:
1426         finish_wait(sk_sleep(sk), &wait);
1427 out:
1428         release_sock(sk);
1429         return err;
1430 }
1431
1432 static int vsock_accept(struct socket *sock, struct socket *newsock, int flags,
1433                         bool kern)
1434 {
1435         struct sock *listener;
1436         int err;
1437         struct sock *connected;
1438         struct vsock_sock *vconnected;
1439         long timeout;
1440         DEFINE_WAIT(wait);
1441
1442         err = 0;
1443         listener = sock->sk;
1444
1445         lock_sock(listener);
1446
1447         if (!sock_type_connectible(sock->type)) {
1448                 err = -EOPNOTSUPP;
1449                 goto out;
1450         }
1451
1452         if (listener->sk_state != TCP_LISTEN) {
1453                 err = -EINVAL;
1454                 goto out;
1455         }
1456
1457         /* Wait for children sockets to appear; these are the new sockets
1458          * created upon connection establishment.
1459          */
1460         timeout = sock_rcvtimeo(listener, flags & O_NONBLOCK);
1461         prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1462
1463         while ((connected = vsock_dequeue_accept(listener)) == NULL &&
1464                listener->sk_err == 0) {
1465                 release_sock(listener);
1466                 timeout = schedule_timeout(timeout);
1467                 finish_wait(sk_sleep(listener), &wait);
1468                 lock_sock(listener);
1469
1470                 if (signal_pending(current)) {
1471                         err = sock_intr_errno(timeout);
1472                         goto out;
1473                 } else if (timeout == 0) {
1474                         err = -EAGAIN;
1475                         goto out;
1476                 }
1477
1478                 prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
1479         }
1480         finish_wait(sk_sleep(listener), &wait);
1481
1482         if (listener->sk_err)
1483                 err = -listener->sk_err;
1484
1485         if (connected) {
1486                 sk_acceptq_removed(listener);
1487
1488                 lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
1489                 vconnected = vsock_sk(connected);
1490
1491                 /* If the listener socket has received an error, then we should
1492                  * reject this socket and return.  Note that we simply mark the
1493                  * socket rejected, drop our reference, and let the cleanup
1494                  * function handle the cleanup; the fact that we found it in
1495                  * the listener's accept queue guarantees that the cleanup
1496                  * function hasn't run yet.
1497                  */
1498                 if (err) {
1499                         vconnected->rejected = true;
1500                 } else {
1501                         newsock->state = SS_CONNECTED;
1502                         sock_graft(connected, newsock);
1503                 }
1504
1505                 release_sock(connected);
1506                 sock_put(connected);
1507         }
1508
1509 out:
1510         release_sock(listener);
1511         return err;
1512 }
1513
1514 static int vsock_listen(struct socket *sock, int backlog)
1515 {
1516         int err;
1517         struct sock *sk;
1518         struct vsock_sock *vsk;
1519
1520         sk = sock->sk;
1521
1522         lock_sock(sk);
1523
1524         if (!sock_type_connectible(sk->sk_type)) {
1525                 err = -EOPNOTSUPP;
1526                 goto out;
1527         }
1528
1529         if (sock->state != SS_UNCONNECTED) {
1530                 err = -EINVAL;
1531                 goto out;
1532         }
1533
1534         vsk = vsock_sk(sk);
1535
1536         if (!vsock_addr_bound(&vsk->local_addr)) {
1537                 err = -EINVAL;
1538                 goto out;
1539         }
1540
1541         sk->sk_max_ack_backlog = backlog;
1542         sk->sk_state = TCP_LISTEN;
1543
1544         err = 0;
1545
1546 out:
1547         release_sock(sk);
1548         return err;
1549 }
1550
1551 static void vsock_update_buffer_size(struct vsock_sock *vsk,
1552                                      const struct vsock_transport *transport,
1553                                      u64 val)
1554 {
1555         if (val > vsk->buffer_max_size)
1556                 val = vsk->buffer_max_size;
1557
1558         if (val < vsk->buffer_min_size)
1559                 val = vsk->buffer_min_size;
1560
1561         if (val != vsk->buffer_size &&
1562             transport && transport->notify_buffer_size)
1563                 transport->notify_buffer_size(vsk, &val);
1564
1565         vsk->buffer_size = val;
1566 }
1567
1568 static int vsock_connectible_setsockopt(struct socket *sock,
1569                                         int level,
1570                                         int optname,
1571                                         sockptr_t optval,
1572                                         unsigned int optlen)
1573 {
1574         int err;
1575         struct sock *sk;
1576         struct vsock_sock *vsk;
1577         const struct vsock_transport *transport;
1578         u64 val;
1579
1580         if (level != AF_VSOCK)
1581                 return -ENOPROTOOPT;
1582
1583 #define COPY_IN(_v)                                       \
1584         do {                                              \
1585                 if (optlen < sizeof(_v)) {                \
1586                         err = -EINVAL;                    \
1587                         goto exit;                        \
1588                 }                                         \
1589                 if (copy_from_sockptr(&_v, optval, sizeof(_v)) != 0) {  \
1590                         err = -EFAULT;                                  \
1591                         goto exit;                                      \
1592                 }                                                       \
1593         } while (0)
1594
1595         err = 0;
1596         sk = sock->sk;
1597         vsk = vsock_sk(sk);
1598
1599         lock_sock(sk);
1600
1601         transport = vsk->transport;
1602
1603         switch (optname) {
1604         case SO_VM_SOCKETS_BUFFER_SIZE:
1605                 COPY_IN(val);
1606                 vsock_update_buffer_size(vsk, transport, val);
1607                 break;
1608
1609         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1610                 COPY_IN(val);
1611                 vsk->buffer_max_size = val;
1612                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1613                 break;
1614
1615         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1616                 COPY_IN(val);
1617                 vsk->buffer_min_size = val;
1618                 vsock_update_buffer_size(vsk, transport, vsk->buffer_size);
1619                 break;
1620
1621         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1622         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD: {
1623                 struct __kernel_sock_timeval tv;
1624
1625                 err = sock_copy_user_timeval(&tv, optval, optlen,
1626                                              optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1627                 if (err)
1628                         break;
1629                 if (tv.tv_sec >= 0 && tv.tv_usec < USEC_PER_SEC &&
1630                     tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) {
1631                         vsk->connect_timeout = tv.tv_sec * HZ +
1632                                 DIV_ROUND_UP((unsigned long)tv.tv_usec, (USEC_PER_SEC / HZ));
1633                         if (vsk->connect_timeout == 0)
1634                                 vsk->connect_timeout =
1635                                     VSOCK_DEFAULT_CONNECT_TIMEOUT;
1636
1637                 } else {
1638                         err = -ERANGE;
1639                 }
1640                 break;
1641         }
1642
1643         default:
1644                 err = -ENOPROTOOPT;
1645                 break;
1646         }
1647
1648 #undef COPY_IN
1649
1650 exit:
1651         release_sock(sk);
1652         return err;
1653 }
1654
1655 static int vsock_connectible_getsockopt(struct socket *sock,
1656                                         int level, int optname,
1657                                         char __user *optval,
1658                                         int __user *optlen)
1659 {
1660         struct sock *sk = sock->sk;
1661         struct vsock_sock *vsk = vsock_sk(sk);
1662
1663         union {
1664                 u64 val64;
1665                 struct old_timeval32 tm32;
1666                 struct __kernel_old_timeval tm;
1667                 struct  __kernel_sock_timeval stm;
1668         } v;
1669
1670         int lv = sizeof(v.val64);
1671         int len;
1672
1673         if (level != AF_VSOCK)
1674                 return -ENOPROTOOPT;
1675
1676         if (get_user(len, optlen))
1677                 return -EFAULT;
1678
1679         memset(&v, 0, sizeof(v));
1680
1681         switch (optname) {
1682         case SO_VM_SOCKETS_BUFFER_SIZE:
1683                 v.val64 = vsk->buffer_size;
1684                 break;
1685
1686         case SO_VM_SOCKETS_BUFFER_MAX_SIZE:
1687                 v.val64 = vsk->buffer_max_size;
1688                 break;
1689
1690         case SO_VM_SOCKETS_BUFFER_MIN_SIZE:
1691                 v.val64 = vsk->buffer_min_size;
1692                 break;
1693
1694         case SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW:
1695         case SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD:
1696                 lv = sock_get_timeout(vsk->connect_timeout, &v,
1697                                       optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
1698                 break;
1699
1700         default:
1701                 return -ENOPROTOOPT;
1702         }
1703
1704         if (len < lv)
1705                 return -EINVAL;
1706         if (len > lv)
1707                 len = lv;
1708         if (copy_to_user(optval, &v, len))
1709                 return -EFAULT;
1710
1711         if (put_user(len, optlen))
1712                 return -EFAULT;
1713
1714         return 0;
1715 }
1716
1717 static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
1718                                      size_t len)
1719 {
1720         struct sock *sk;
1721         struct vsock_sock *vsk;
1722         const struct vsock_transport *transport;
1723         ssize_t total_written;
1724         long timeout;
1725         int err;
1726         struct vsock_transport_send_notify_data send_data;
1727         DEFINE_WAIT_FUNC(wait, woken_wake_function);
1728
1729         sk = sock->sk;
1730         vsk = vsock_sk(sk);
1731         total_written = 0;
1732         err = 0;
1733
1734         if (msg->msg_flags & MSG_OOB)
1735                 return -EOPNOTSUPP;
1736
1737         lock_sock(sk);
1738
1739         transport = vsk->transport;
1740
1741         /* Callers should not provide a destination with connection oriented
1742          * sockets.
1743          */
1744         if (msg->msg_namelen) {
1745                 err = sk->sk_state == TCP_ESTABLISHED ? -EISCONN : -EOPNOTSUPP;
1746                 goto out;
1747         }
1748
1749         /* Send data only if both sides are not shutdown in the direction. */
1750         if (sk->sk_shutdown & SEND_SHUTDOWN ||
1751             vsk->peer_shutdown & RCV_SHUTDOWN) {
1752                 err = -EPIPE;
1753                 goto out;
1754         }
1755
1756         if (!transport || sk->sk_state != TCP_ESTABLISHED ||
1757             !vsock_addr_bound(&vsk->local_addr)) {
1758                 err = -ENOTCONN;
1759                 goto out;
1760         }
1761
1762         if (!vsock_addr_bound(&vsk->remote_addr)) {
1763                 err = -EDESTADDRREQ;
1764                 goto out;
1765         }
1766
1767         /* Wait for room in the produce queue to enqueue our user's data. */
1768         timeout = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
1769
1770         err = transport->notify_send_init(vsk, &send_data);
1771         if (err < 0)
1772                 goto out;
1773
1774         while (total_written < len) {
1775                 ssize_t written;
1776
1777                 add_wait_queue(sk_sleep(sk), &wait);
1778                 while (vsock_stream_has_space(vsk) == 0 &&
1779                        sk->sk_err == 0 &&
1780                        !(sk->sk_shutdown & SEND_SHUTDOWN) &&
1781                        !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
1782
1783                         /* Don't wait for non-blocking sockets. */
1784                         if (timeout == 0) {
1785                                 err = -EAGAIN;
1786                                 remove_wait_queue(sk_sleep(sk), &wait);
1787                                 goto out_err;
1788                         }
1789
1790                         err = transport->notify_send_pre_block(vsk, &send_data);
1791                         if (err < 0) {
1792                                 remove_wait_queue(sk_sleep(sk), &wait);
1793                                 goto out_err;
1794                         }
1795
1796                         release_sock(sk);
1797                         timeout = wait_woken(&wait, TASK_INTERRUPTIBLE, timeout);
1798                         lock_sock(sk);
1799                         if (signal_pending(current)) {
1800                                 err = sock_intr_errno(timeout);
1801                                 remove_wait_queue(sk_sleep(sk), &wait);
1802                                 goto out_err;
1803                         } else if (timeout == 0) {
1804                                 err = -EAGAIN;
1805                                 remove_wait_queue(sk_sleep(sk), &wait);
1806                                 goto out_err;
1807                         }
1808                 }
1809                 remove_wait_queue(sk_sleep(sk), &wait);
1810
1811                 /* These checks occur both as part of and after the loop
1812                  * conditional since we need to check before and after
1813                  * sleeping.
1814                  */
1815                 if (sk->sk_err) {
1816                         err = -sk->sk_err;
1817                         goto out_err;
1818                 } else if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1819                            (vsk->peer_shutdown & RCV_SHUTDOWN)) {
1820                         err = -EPIPE;
1821                         goto out_err;
1822                 }
1823
1824                 err = transport->notify_send_pre_enqueue(vsk, &send_data);
1825                 if (err < 0)
1826                         goto out_err;
1827
1828                 /* Note that enqueue will only write as many bytes as are free
1829                  * in the produce queue, so we don't need to ensure len is
1830                  * smaller than the queue size.  It is the caller's
1831                  * responsibility to check how many bytes we were able to send.
1832                  */
1833
1834                 if (sk->sk_type == SOCK_SEQPACKET) {
1835                         written = transport->seqpacket_enqueue(vsk,
1836                                                 msg, len - total_written);
1837                 } else {
1838                         written = transport->stream_enqueue(vsk,
1839                                         msg, len - total_written);
1840                 }
1841                 if (written < 0) {
1842                         err = -ENOMEM;
1843                         goto out_err;
1844                 }
1845
1846                 total_written += written;
1847
1848                 err = transport->notify_send_post_enqueue(
1849                                 vsk, written, &send_data);
1850                 if (err < 0)
1851                         goto out_err;
1852
1853         }
1854
1855 out_err:
1856         if (total_written > 0) {
1857                 /* Return number of written bytes only if:
1858                  * 1) SOCK_STREAM socket.
1859                  * 2) SOCK_SEQPACKET socket when whole buffer is sent.
1860                  */
1861                 if (sk->sk_type == SOCK_STREAM || total_written == len)
1862                         err = total_written;
1863         }
1864 out:
1865         release_sock(sk);
1866         return err;
1867 }
1868
1869 static int vsock_connectible_wait_data(struct sock *sk,
1870                                        struct wait_queue_entry *wait,
1871                                        long timeout,
1872                                        struct vsock_transport_recv_notify_data *recv_data,
1873                                        size_t target)
1874 {
1875         const struct vsock_transport *transport;
1876         struct vsock_sock *vsk;
1877         s64 data;
1878         int err;
1879
1880         vsk = vsock_sk(sk);
1881         err = 0;
1882         transport = vsk->transport;
1883
1884         while ((data = vsock_connectible_has_data(vsk)) == 0) {
1885                 prepare_to_wait(sk_sleep(sk), wait, TASK_INTERRUPTIBLE);
1886
1887                 if (sk->sk_err != 0 ||
1888                     (sk->sk_shutdown & RCV_SHUTDOWN) ||
1889                     (vsk->peer_shutdown & SEND_SHUTDOWN)) {
1890                         break;
1891                 }
1892
1893                 /* Don't wait for non-blocking sockets. */
1894                 if (timeout == 0) {
1895                         err = -EAGAIN;
1896                         break;
1897                 }
1898
1899                 if (recv_data) {
1900                         err = transport->notify_recv_pre_block(vsk, target, recv_data);
1901                         if (err < 0)
1902                                 break;
1903                 }
1904
1905                 release_sock(sk);
1906                 timeout = schedule_timeout(timeout);
1907                 lock_sock(sk);
1908
1909                 if (signal_pending(current)) {
1910                         err = sock_intr_errno(timeout);
1911                         break;
1912                 } else if (timeout == 0) {
1913                         err = -EAGAIN;
1914                         break;
1915                 }
1916         }
1917
1918         finish_wait(sk_sleep(sk), wait);
1919
1920         if (err)
1921                 return err;
1922
1923         /* Internal transport error when checking for available
1924          * data. XXX This should be changed to a connection
1925          * reset in a later change.
1926          */
1927         if (data < 0)
1928                 return -ENOMEM;
1929
1930         return data;
1931 }
1932
1933 static int __vsock_stream_recvmsg(struct sock *sk, struct msghdr *msg,
1934                                   size_t len, int flags)
1935 {
1936         struct vsock_transport_recv_notify_data recv_data;
1937         const struct vsock_transport *transport;
1938         struct vsock_sock *vsk;
1939         ssize_t copied;
1940         size_t target;
1941         long timeout;
1942         int err;
1943
1944         DEFINE_WAIT(wait);
1945
1946         vsk = vsock_sk(sk);
1947         transport = vsk->transport;
1948
1949         /* We must not copy less than target bytes into the user's buffer
1950          * before returning successfully, so we wait for the consume queue to
1951          * have that much data to consume before dequeueing.  Note that this
1952          * makes it impossible to handle cases where target is greater than the
1953          * queue size.
1954          */
1955         target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1956         if (target >= transport->stream_rcvhiwat(vsk)) {
1957                 err = -ENOMEM;
1958                 goto out;
1959         }
1960         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1961         copied = 0;
1962
1963         err = transport->notify_recv_init(vsk, target, &recv_data);
1964         if (err < 0)
1965                 goto out;
1966
1967
1968         while (1) {
1969                 ssize_t read;
1970
1971                 err = vsock_connectible_wait_data(sk, &wait, timeout,
1972                                                   &recv_data, target);
1973                 if (err <= 0)
1974                         break;
1975
1976                 err = transport->notify_recv_pre_dequeue(vsk, target,
1977                                                          &recv_data);
1978                 if (err < 0)
1979                         break;
1980
1981                 read = transport->stream_dequeue(vsk, msg, len - copied, flags);
1982                 if (read < 0) {
1983                         err = -ENOMEM;
1984                         break;
1985                 }
1986
1987                 copied += read;
1988
1989                 err = transport->notify_recv_post_dequeue(vsk, target, read,
1990                                                 !(flags & MSG_PEEK), &recv_data);
1991                 if (err < 0)
1992                         goto out;
1993
1994                 if (read >= target || flags & MSG_PEEK)
1995                         break;
1996
1997                 target -= read;
1998         }
1999
2000         if (sk->sk_err)
2001                 err = -sk->sk_err;
2002         else if (sk->sk_shutdown & RCV_SHUTDOWN)
2003                 err = 0;
2004
2005         if (copied > 0)
2006                 err = copied;
2007
2008 out:
2009         return err;
2010 }
2011
2012 static int __vsock_seqpacket_recvmsg(struct sock *sk, struct msghdr *msg,
2013                                      size_t len, int flags)
2014 {
2015         const struct vsock_transport *transport;
2016         struct vsock_sock *vsk;
2017         ssize_t msg_len;
2018         long timeout;
2019         int err = 0;
2020         DEFINE_WAIT(wait);
2021
2022         vsk = vsock_sk(sk);
2023         transport = vsk->transport;
2024
2025         timeout = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
2026
2027         err = vsock_connectible_wait_data(sk, &wait, timeout, NULL, 0);
2028         if (err <= 0)
2029                 goto out;
2030
2031         msg_len = transport->seqpacket_dequeue(vsk, msg, flags);
2032
2033         if (msg_len < 0) {
2034                 err = -ENOMEM;
2035                 goto out;
2036         }
2037
2038         if (sk->sk_err) {
2039                 err = -sk->sk_err;
2040         } else if (sk->sk_shutdown & RCV_SHUTDOWN) {
2041                 err = 0;
2042         } else {
2043                 /* User sets MSG_TRUNC, so return real length of
2044                  * packet.
2045                  */
2046                 if (flags & MSG_TRUNC)
2047                         err = msg_len;
2048                 else
2049                         err = len - msg_data_left(msg);
2050
2051                 /* Always set MSG_TRUNC if real length of packet is
2052                  * bigger than user's buffer.
2053                  */
2054                 if (msg_len > len)
2055                         msg->msg_flags |= MSG_TRUNC;
2056         }
2057
2058 out:
2059         return err;
2060 }
2061
2062 static int
2063 vsock_connectible_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2064                           int flags)
2065 {
2066         struct sock *sk;
2067         struct vsock_sock *vsk;
2068         const struct vsock_transport *transport;
2069         int err;
2070
2071         DEFINE_WAIT(wait);
2072
2073         sk = sock->sk;
2074         vsk = vsock_sk(sk);
2075         err = 0;
2076
2077         lock_sock(sk);
2078
2079         transport = vsk->transport;
2080
2081         if (!transport || sk->sk_state != TCP_ESTABLISHED) {
2082                 /* Recvmsg is supposed to return 0 if a peer performs an
2083                  * orderly shutdown. Differentiate between that case and when a
2084                  * peer has not connected or a local shutdown occurred with the
2085                  * SOCK_DONE flag.
2086                  */
2087                 if (sock_flag(sk, SOCK_DONE))
2088                         err = 0;
2089                 else
2090                         err = -ENOTCONN;
2091
2092                 goto out;
2093         }
2094
2095         if (flags & MSG_OOB) {
2096                 err = -EOPNOTSUPP;
2097                 goto out;
2098         }
2099
2100         /* We don't check peer_shutdown flag here since peer may actually shut
2101          * down, but there can be data in the queue that a local socket can
2102          * receive.
2103          */
2104         if (sk->sk_shutdown & RCV_SHUTDOWN) {
2105                 err = 0;
2106                 goto out;
2107         }
2108
2109         /* It is valid on Linux to pass in a zero-length receive buffer.  This
2110          * is not an error.  We may as well bail out now.
2111          */
2112         if (!len) {
2113                 err = 0;
2114                 goto out;
2115         }
2116
2117         if (sk->sk_type == SOCK_STREAM)
2118                 err = __vsock_stream_recvmsg(sk, msg, len, flags);
2119         else
2120                 err = __vsock_seqpacket_recvmsg(sk, msg, len, flags);
2121
2122 out:
2123         release_sock(sk);
2124         return err;
2125 }
2126
2127 static const struct proto_ops vsock_stream_ops = {
2128         .family = PF_VSOCK,
2129         .owner = THIS_MODULE,
2130         .release = vsock_release,
2131         .bind = vsock_bind,
2132         .connect = vsock_connect,
2133         .socketpair = sock_no_socketpair,
2134         .accept = vsock_accept,
2135         .getname = vsock_getname,
2136         .poll = vsock_poll,
2137         .ioctl = sock_no_ioctl,
2138         .listen = vsock_listen,
2139         .shutdown = vsock_shutdown,
2140         .setsockopt = vsock_connectible_setsockopt,
2141         .getsockopt = vsock_connectible_getsockopt,
2142         .sendmsg = vsock_connectible_sendmsg,
2143         .recvmsg = vsock_connectible_recvmsg,
2144         .mmap = sock_no_mmap,
2145         .sendpage = sock_no_sendpage,
2146 };
2147
2148 static const struct proto_ops vsock_seqpacket_ops = {
2149         .family = PF_VSOCK,
2150         .owner = THIS_MODULE,
2151         .release = vsock_release,
2152         .bind = vsock_bind,
2153         .connect = vsock_connect,
2154         .socketpair = sock_no_socketpair,
2155         .accept = vsock_accept,
2156         .getname = vsock_getname,
2157         .poll = vsock_poll,
2158         .ioctl = sock_no_ioctl,
2159         .listen = vsock_listen,
2160         .shutdown = vsock_shutdown,
2161         .setsockopt = vsock_connectible_setsockopt,
2162         .getsockopt = vsock_connectible_getsockopt,
2163         .sendmsg = vsock_connectible_sendmsg,
2164         .recvmsg = vsock_connectible_recvmsg,
2165         .mmap = sock_no_mmap,
2166         .sendpage = sock_no_sendpage,
2167 };
2168
2169 static int vsock_create(struct net *net, struct socket *sock,
2170                         int protocol, int kern)
2171 {
2172         struct vsock_sock *vsk;
2173         struct sock *sk;
2174         int ret;
2175
2176         if (!sock)
2177                 return -EINVAL;
2178
2179         if (protocol && protocol != PF_VSOCK)
2180                 return -EPROTONOSUPPORT;
2181
2182         switch (sock->type) {
2183         case SOCK_DGRAM:
2184                 sock->ops = &vsock_dgram_ops;
2185                 break;
2186         case SOCK_STREAM:
2187                 sock->ops = &vsock_stream_ops;
2188                 break;
2189         case SOCK_SEQPACKET:
2190                 sock->ops = &vsock_seqpacket_ops;
2191                 break;
2192         default:
2193                 return -ESOCKTNOSUPPORT;
2194         }
2195
2196         sock->state = SS_UNCONNECTED;
2197
2198         sk = __vsock_create(net, sock, NULL, GFP_KERNEL, 0, kern);
2199         if (!sk)
2200                 return -ENOMEM;
2201
2202         vsk = vsock_sk(sk);
2203
2204         if (sock->type == SOCK_DGRAM) {
2205                 ret = vsock_assign_transport(vsk, NULL);
2206                 if (ret < 0) {
2207                         sock_put(sk);
2208                         return ret;
2209                 }
2210         }
2211
2212         vsock_insert_unbound(vsk);
2213
2214         return 0;
2215 }
2216
2217 static const struct net_proto_family vsock_family_ops = {
2218         .family = AF_VSOCK,
2219         .create = vsock_create,
2220         .owner = THIS_MODULE,
2221 };
2222
2223 static long vsock_dev_do_ioctl(struct file *filp,
2224                                unsigned int cmd, void __user *ptr)
2225 {
2226         u32 __user *p = ptr;
2227         u32 cid = VMADDR_CID_ANY;
2228         int retval = 0;
2229
2230         switch (cmd) {
2231         case IOCTL_VM_SOCKETS_GET_LOCAL_CID:
2232                 /* To be compatible with the VMCI behavior, we prioritize the
2233                  * guest CID instead of well-know host CID (VMADDR_CID_HOST).
2234                  */
2235                 if (transport_g2h)
2236                         cid = transport_g2h->get_local_cid();
2237                 else if (transport_h2g)
2238                         cid = transport_h2g->get_local_cid();
2239
2240                 if (put_user(cid, p) != 0)
2241                         retval = -EFAULT;
2242                 break;
2243
2244         default:
2245                 retval = -ENOIOCTLCMD;
2246         }
2247
2248         return retval;
2249 }
2250
2251 static long vsock_dev_ioctl(struct file *filp,
2252                             unsigned int cmd, unsigned long arg)
2253 {
2254         return vsock_dev_do_ioctl(filp, cmd, (void __user *)arg);
2255 }
2256
2257 #ifdef CONFIG_COMPAT
2258 static long vsock_dev_compat_ioctl(struct file *filp,
2259                                    unsigned int cmd, unsigned long arg)
2260 {
2261         return vsock_dev_do_ioctl(filp, cmd, compat_ptr(arg));
2262 }
2263 #endif
2264
2265 static const struct file_operations vsock_device_ops = {
2266         .owner          = THIS_MODULE,
2267         .unlocked_ioctl = vsock_dev_ioctl,
2268 #ifdef CONFIG_COMPAT
2269         .compat_ioctl   = vsock_dev_compat_ioctl,
2270 #endif
2271         .open           = nonseekable_open,
2272 };
2273
2274 static struct miscdevice vsock_device = {
2275         .name           = "vsock",
2276         .fops           = &vsock_device_ops,
2277 };
2278
2279 static int __init vsock_init(void)
2280 {
2281         int err = 0;
2282
2283         vsock_init_tables();
2284
2285         vsock_proto.owner = THIS_MODULE;
2286         vsock_device.minor = MISC_DYNAMIC_MINOR;
2287         err = misc_register(&vsock_device);
2288         if (err) {
2289                 pr_err("Failed to register misc device\n");
2290                 goto err_reset_transport;
2291         }
2292
2293         err = proto_register(&vsock_proto, 1);  /* we want our slab */
2294         if (err) {
2295                 pr_err("Cannot register vsock protocol\n");
2296                 goto err_deregister_misc;
2297         }
2298
2299         err = sock_register(&vsock_family_ops);
2300         if (err) {
2301                 pr_err("could not register af_vsock (%d) address family: %d\n",
2302                        AF_VSOCK, err);
2303                 goto err_unregister_proto;
2304         }
2305
2306         return 0;
2307
2308 err_unregister_proto:
2309         proto_unregister(&vsock_proto);
2310 err_deregister_misc:
2311         misc_deregister(&vsock_device);
2312 err_reset_transport:
2313         return err;
2314 }
2315
2316 static void __exit vsock_exit(void)
2317 {
2318         misc_deregister(&vsock_device);
2319         sock_unregister(AF_VSOCK);
2320         proto_unregister(&vsock_proto);
2321 }
2322
2323 const struct vsock_transport *vsock_core_get_transport(struct vsock_sock *vsk)
2324 {
2325         return vsk->transport;
2326 }
2327 EXPORT_SYMBOL_GPL(vsock_core_get_transport);
2328
2329 int vsock_core_register(const struct vsock_transport *t, int features)
2330 {
2331         const struct vsock_transport *t_h2g, *t_g2h, *t_dgram, *t_local;
2332         int err = mutex_lock_interruptible(&vsock_register_mutex);
2333
2334         if (err)
2335                 return err;
2336
2337         t_h2g = transport_h2g;
2338         t_g2h = transport_g2h;
2339         t_dgram = transport_dgram;
2340         t_local = transport_local;
2341
2342         if (features & VSOCK_TRANSPORT_F_H2G) {
2343                 if (t_h2g) {
2344                         err = -EBUSY;
2345                         goto err_busy;
2346                 }
2347                 t_h2g = t;
2348         }
2349
2350         if (features & VSOCK_TRANSPORT_F_G2H) {
2351                 if (t_g2h) {
2352                         err = -EBUSY;
2353                         goto err_busy;
2354                 }
2355                 t_g2h = t;
2356         }
2357
2358         if (features & VSOCK_TRANSPORT_F_DGRAM) {
2359                 if (t_dgram) {
2360                         err = -EBUSY;
2361                         goto err_busy;
2362                 }
2363                 t_dgram = t;
2364         }
2365
2366         if (features & VSOCK_TRANSPORT_F_LOCAL) {
2367                 if (t_local) {
2368                         err = -EBUSY;
2369                         goto err_busy;
2370                 }
2371                 t_local = t;
2372         }
2373
2374         transport_h2g = t_h2g;
2375         transport_g2h = t_g2h;
2376         transport_dgram = t_dgram;
2377         transport_local = t_local;
2378
2379 err_busy:
2380         mutex_unlock(&vsock_register_mutex);
2381         return err;
2382 }
2383 EXPORT_SYMBOL_GPL(vsock_core_register);
2384
2385 void vsock_core_unregister(const struct vsock_transport *t)
2386 {
2387         mutex_lock(&vsock_register_mutex);
2388
2389         if (transport_h2g == t)
2390                 transport_h2g = NULL;
2391
2392         if (transport_g2h == t)
2393                 transport_g2h = NULL;
2394
2395         if (transport_dgram == t)
2396                 transport_dgram = NULL;
2397
2398         if (transport_local == t)
2399                 transport_local = NULL;
2400
2401         mutex_unlock(&vsock_register_mutex);
2402 }
2403 EXPORT_SYMBOL_GPL(vsock_core_unregister);
2404
2405 module_init(vsock_init);
2406 module_exit(vsock_exit);
2407
2408 MODULE_AUTHOR("VMware, Inc.");
2409 MODULE_DESCRIPTION("VMware Virtual Socket Family");
2410 MODULE_VERSION("1.0.2.0-k");
2411 MODULE_LICENSE("GPL v2");